Hello Everyone
Rather than using these pinned threads for shameless self promotion all of the time I thought it would be a nice change to rotate in some educational information on various hosting topics. I’m not writing any of this to instigate arguments or comparisons between specific hosts. Rather, to give some quality information and my personal view on the how and why we do things. To those of you that take the time to read this, thank you in advance and I hope you enjoy the information below.
The basic concept that I wanted to share today is one of routing performance and bandwidth quality. As you know, virtually all hosting companies connect to at least one but usually multiple global networks. The average host will advertise that they are fully redundant and use BGP4 to communicate with all of their carriers. I am expressing that it is my professional opinion that this more ‘traditional’ setup is inadequate and that even hosts advertising “premium bandwidth” will have less than ideal routing unless they utilize the right routing tools and strategies.
For the sake of brevity, this is not an article on whole network setup, not even about specific carriers, even though I will be referencing some for data purposes. My purpose is to express the following opinion and then back it up with factual references so that you as a host consumer can make more educated decisions about what is best for your business.
Brad Says: No single global carrier is “the best” all of the time. All carriers regardless size, retail cost of bandwidth and market reputation experience network congestion, packet loss, router problems, downtime, peering problems and poor route decision making.
In a more traditional setup hosts are given “routing tables” on a regular basis from each of the carriers that they connect to. The host routers presume the information to be valid and then subsequently divide traffic across the different carriers they connect to based on the volumes they want to see on each of their links. In other words, it’s not inherently a performance based decision or one that checks the validity of the routes being passed from their carriers selling them bandwidth.
On MojoHost’s network we deploy a device manufactured by a company called Internap called the Flow Control Platform (FCP). I believe that the end result of having this additional piece of equipment on our network is substantially improved bandwidth performance and more intelligent routing across all of the 7 carriers and 34 gigabit of current connectivity. More on that in a little bit…
My basic belief is that as a host we must assume that every network is not perfect and neither is the route information they pass along to us. Every network in our opinion has both strong and weak components that are ever changing in nature. Manually doing the work of checking routes for packet loss, jitter and latency is impossible. Assuming that the route paths given to us by carriers are correct, working and the best possible path would be flawed and in many instances incorrect.
The device that I deploy on our network earlier referred to as the FCP does some great things. In a nutshell, it is constantly performance checking and rebuilding our MojoHost routing table to the rest of the world. It takes the top 50,000 destination networks surfing our network and does route analysis across all 7 carriers we connect to on a constant basis to develop our end state routing table. On an average day it makes 50,000 route changes to and from various networks, all on auto-pilot, just based on performance decisions.
I believe the proof is in the results. I think that by using a device like this that the end result is our intelligently blended bandwidth product is better than most competitors and clearly makes the case that no one network provider has the best routes all of the time… that the best case scenario is intelligent monitoring and bandwidth distribution across a wide variety of global carriers.
We can all agree that Level3 has a superb reputation in the marketplace as they do have an expansive, high quality global footprint and operation. During a one hour period this is a list of round 450 performance based changes that our FCP made to take traffic off of our Level3 link and place it on other networks. During the 60 minute period our network device made around 2200 route changes to and from various carriers. Below are links to both the l3 route changes and the whole list. If you look at them what you want to focus on is the round-trip time on the old path and the round-trip time on the new path in addition to any changes in packet loss that are identified between old and new carrier selection.
Short text list summary of changes from Level3 to other carriers:
http://www.mojohost.com/gfy/level3.txt
Excel spreadsheet showing 2228 changes across multiple carriers:
http://www.mojohost.com/gfy/ControlEvents.xls
The more grand point that I am trying to illustrate is that the whole of network planning and implementation is often more important and sophisticated (and expensive) than simply connecting to a bunch of carriers or just buying bandwidth from a ‘marquis’ value carrier. What we clearly see on a daily basis is that many of these networks have substantial value and performance to offer and that the best practice is to implement as many as possible and then intelligently monitor and change outbound traffic routes based on solid performance data.
If anyone has any questions, feel free to chat me up here, via email, ICQ or phone.
Sincerely,
Brad Mitchell
P.S. - Those New Years specials I posted last week are still good ;)